DEBUGGING EFFORT ESTIMATION USING SOFTWARE METRICS

Measurements of 23 style characteristics, and the program metrics LOC, V(g), VARS, and PARS were collected from student Cobol programs by a program analyzer. These measurements, together with debugging time (syntax and logic) data, were analyzed using several statistical procedures of SAS, including linear, quadratic, and multiple regressions. Some of the characteristics shown to significantly correlate with debug time are GOTO usage, structuring of the IF-ELSE construct, level 88 item usage, paragraph invocation pattern, and data name length. Among the observed characteristic measures which are associated with lowest debug times are: 17 percent blank lines in the Data Division, 12 percent blank lines in the Procedure Division, and 13 character long data items. A debugging effort estimator, DEST, was developed to estimate debug times. This estimator, a quadratic function of nine characteristics, has a coefficient of multiple determination (R2) of 0.7551 with the total debug time (significance level 0.0001). None of the software metrics LOC, V(g), VARS, and PARS has r2 values greater than 0.3 when regressed with total debug time. The variables of DEST, when regressed with debug times from various subsets of the programs stratified by LOC, V(.g), and student GPA, had high R2 values. © 1990 IEEE